`@anvia/azure`

The Azure provider adapter connects Anvia agents to Azure OpenAI and Azure AI Foundry endpoints. It owns Azure endpoint and credential configuration, provider identity, and Azure Responses stream normalization, and reuses @anvia/openai for request and response mapping.


Support	First-party
Version	`0.1.0`
Runtime	ESM, server-side JavaScript
Peer	Matching `@anvia/core` stable release
Depends on	`@anvia/openai`, official `openai` SDK

Install

bash

pnpm add @anvia/azure @anvia/core

Create an Azure agent

import { Agent } from '@anvia/core'
import { AzureOpenAIClient } from '@anvia/azure'

const client = new AzureOpenAIClient({
  endpoint: process.env.AZURE_OPENAI_ENDPOINT!,
  apiKey: process.env.AZURE_OPENAI_API_KEY!,
})

const agent = new Agent({
  id: 'assistant',
  model: client.completionModel({
    modelId: process.env.AZURE_OPENAI_DEPLOYMENT!,
    api: 'responses',
  }),
})

const result = await agent.generate({ prompt: 'Hello!' })

if (result.type === 'response') {
  console.log(result.output)
}

endpoint is the resource origin, such as https://example.openai.azure.com; the client appends /openai/v1/. Use baseUrl instead for a full API URL, including a Foundry project endpoint. modelId is your Azure deployment name, not necessarily the underlying model name. api is optional and defaults to Chat Completions; pass api: 'responses' for the Responses API.

Capabilities

Capability	Factory	Default
Streaming completion	`completionModel({ modelId, api? })`	Deployment name; Chat Completions
Dense embeddings	`embeddingModel({ modelId })`	Deployment name
Image generation	`imageGenerationModel({ modelId })`	Deployment name
Text-to-speech	`speechGenerationModel({ modelId })`	Deployment name
Transcription	`transcriptionModel({ modelId })`	Deployment name
Model inventory	`listModels()`	Endpoint model list

Availability of APIs, tools, media options, and model listing depends on your Azure endpoint and deployment. Handles and listing errors identify the provider as azure-openai; request traces use azure-openai-chat or azure-openai-responses.

Authentication

Pass exactly one credential:

apiKey, a non-empty Azure key.
azureADTokenProvider, an asynchronous function that returns a Microsoft Entra bearer token, such as the result of getBearerTokenProvider from @azure/identity.

Alternatively, inject an existing OpenAI SDK client with client, for example the SDK's AzureOpenAI client for versioned endpoints that need api-version. See Configuration.

Compatibility

@anvia/azure is an ESM package. The managed client targets the Azure v1 API, which does not require an api-version query parameter, and disables SDK retries so Anvia owns retry policy. Azure Responses requests tag conversation messages with type: "message" for Foundry project endpoints, and the Azure Responses stream normalizer resolves a function name that Azure omits from the terminal tool-argument event.

The package does not select credentials or endpoints from environment variables, enumerate Azure Resource Manager deployments, or provide AI SDK-style file-ID prefix inference, hosted-tool helpers, or Azure Speech and MAI adapters.

@anvia/azure ​

Install ​

Create an Azure agent ​

Capabilities ​

Authentication ​

Compatibility ​